# Security and Memory-Governance Notice

NZFC-GRAM treats retrieved memory as evidence, not instruction.

Security-relevant mechanisms include:

- source/authority metadata for memory cards,
- instruction_allowed flags,
- trust_level scores,
- redaction of untrusted prompt-injection-like memory before model insertion,
- context-governed memory-pack generation,
- no insertion of the full 10M archive into model context,
- answer-side internal-memory-claim checks,
- claim-evidence verification,
- scope isolation by user/project/session,
- tombstone/deleted-memory filtering.

This repository is a research/runtime release and is not a formal production security certification.

Do not store secrets, credentials, private keys, regulated personal data, or confidential customer data in the local memory store unless you have implemented the necessary security, encryption, access-control, audit, and retention policies.